Husai Releases Kosmo: A Data-Driven Approach to Embodied AI
禾赛造“相机”：死磕具身数据的激光雷达公司才是好机器人公司
Husai, a LiDAR company, has launched Kosmo, a 3D world data engine that captures real environments and converts them into editable, reusable data for robot training, world modeling, and content creation. The system includes the Kosmo One AI space camera and the Kosmo World cloud platform, enabling efficient spatial reconstruction and scene modification for diverse applications.
Husai's Kosmo represents a shift from traditional LiDAR to a data-centric approach for embodied AI, offering tools for real-world data collection and simulation integration. Its focus on spatial reconstruction and reusability addresses challenges in training robots and content creation.
Source: Chinese robotics — Dataset and collection discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
training_steps
55
View original evidence
据 Sharpa 介绍，其机器人在上海一家 DQ 门店使用现有设备制作暴风雪冰淇淋，一单包含 55 个连续操作步骤，涉及开柜门、添加配料、搅拌和翻转纸杯等动作。
Open source S19elapsed_hours
300 hours
View original evidence
在这一项目中，原本约需专业团队 300 小时完成的训练环境建模，被约 15 分钟的扫描与重建替代；该部署中的机器人任务成功率达到 99%，失败率降至此前的十分之一。
Open source S20Reported duration
≈15 minutes
View original evidence
在这一项目中，原本约需专业团队 300 小时完成的训练环境建模，被约 15 分钟的扫描与重建替代；该部署中的机器人任务成功率达到 99%，失败率降至此前的十分之一。
Open source S20Reported success rate
99%
View original evidence
在这一项目中，原本约需专业团队 300 小时完成的训练环境建模，被约 15 分钟的扫描与重建替代；该部署中的机器人任务成功率达到 99%，失败率降至此前的十分之一。
Open source S20model_generation_speed
5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Dataset licensing and commercial-use rights are not established here.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
据 Sharpa 介绍，其机器人在上海一家 DQ 门店使用现有设备制作暴风雪冰淇淋，一单包含 55 个连续操作步骤，涉及开柜门、添加配料、搅拌和翻转纸杯等动作。
Open source S19
在这一项目中，原本约需专业团队 300 小时完成的训练环境建模，被约 15 分钟的扫描与重建替代；该部署中的机器人任务成功率达到 99%，失败率降至此前的十分之一。
Open source S20
云端模型生成速度，在相近硬件、数据集和质量要求下，最高约为同类 3DGS 重建产品的 5 倍。
Open source S14
对机器人训练来说，这意味着一次采集可以扩展出多种训练条件。
Open source S15
同一项拿杯子或收拾桌面的任务，可以在不同条件下反复训练，帮助机器人学习适应环境变化。
Open source S17
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：Chinese robotics — Dataset and collection discovery · sohu.com